Qwen 3.5:4b hits 40 t/s on legacy GPUs

The efficiency of small-parameter models like Qwen 3.5 is reaching a tipping point where legacy consumer hardware can rival cloud-based coding assistants. Performance on 8GB VRAM validates the industry shift toward optimized models for specialized developer tasks, and integration with Qdrant for RAG effectively overcomes context limitations. Achieving 40 tokens/sec on an eight-year-old GTX 1070 Ti indicates that a private, zero-latency coding experience is now accessible locally without heavy hardware investment.